The primary motivation of this dissertation is to investigate how to enable interoperability in the logistics domain by the aid of ontology alignment. More in detail, the primary research objective of this dissertation is To address interoperability between heterogeneous IT systems in logistics by using ontology alignment. To accomplish the objective, we first look into the literature of ontology alignment using a quantitative approach to get a thorough understanding of the available literature and its progress. We particularly identify several research gaps that are studied in the subsequent chapters that serve the objective of this dissertation. An important lesson learned from the literature analysis is that there are two segregated communities that form ontology alignment but work independently and with minimal interactions with each other. Based on the identified research gaps, we develop a new system, SANOM (simulated annealing-based ontology matching), that addresses the non-deterministic polynomialtime( NP) ontology alignment problembased on thewell-knownevolutionary algorithm, simulated annealing. SANOMis equippedwith an extended Soft TF-IDF (termfrequencyinverse document frequency) string similarity metric that can also detect linguistic similarity among the names of entities in two ontologies in question. Structural similarity metrics are also taken into account that increase the alignment performance for more complex ontologies. Simulated annealing with a warm initialization is used as the matching strategy to find an optimal solution to the ontology alignment problem. The experiments show that SANOM has a very competitive performance with the best systems that participated in the ontology alignment evaluation initiative (OAEI) and is particularly faster than other evolutionary algorithm-based alignment systems. To come to a better understanding of which alignment system is preferred, we develop several methods for evaluation and comparison of alignment systems using different statistical techniques and multi-criteria decision-making methods (MCDM). We first study the frequentist approach for comparing alignment systems. More in detail, we compare different statistical tests for comparing alignment systems over single or multiple benchmarks and propose a proper test based on the number of benchmarks and alignment systems. While these techniques are more reliable than those being currently in use, it suffers from the drawbacks of making decisions based on p-values, which can be addressed by Bayesian statistics.
|Qualification||Doctor of Philosophy|
|Award date||28 Feb 2020|
|Publication status||Published - 2020|
- Ontology alignment
- simulated annealing